一种消除椒盐噪声的迭代自适应加权均值滤波  

An Iterative Adaptive Weighted Mean Filter for Removing Salt and Pepper Noise

作　　者：秦承刚 张建[1] 施赣明 张学阁 谢敬心 徐世许[3] QIN Chenggang;ZHANG Jian;SHI Ganming;ZHANG Xuege;XIE Jingxin;XU Shixu(China Tobacco Shandong Industrial Co.,Ltd.Qingdao Cigarette Factory,Qingdao 266101,China;China Tobacco Shandong Industrial Co.,Ltd.,Jinan 250014,China;School of Automation,Qingdao University,Qingdao 266071,China)

机构地区：[1]山东中烟工业有限责任公司青岛卷烟厂,山东青岛266101 [2]山东中烟工业有限责任公司,山东济南250014 [3]青岛大学自动化学院,山东青岛266071

出　　处：《青岛大学学报（工程技术版）》2024年第3期39-45,共7页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：山东中烟工业有限责任公司科技基金资助项目(202201025)。

摘　　要：针对传统的均值滤波和多种中值滤波不能有效去除高密度椒盐噪声的不足,提出一种迭代自适应加权均值滤波算法。为了能够对图像外围的像素点进行滤波,沿四周进行图像扩展。根据椒盐噪声的极值特性,将像素点分为噪声点和信号点,噪声点采用多轮迭代,利用前次滤波的结果,求取滤波窗口非噪声点灰度值的加权均值,基于与窗口中心像素的空间距离计算加权系数;信号点保持原值。滤波窗口尺寸根据窗口内是否含有非噪声点自适应地由小变大。选用5种不同的算法在噪声密度20%~80%范围内进行仿真对比,验证算法的有效性。仿真结果表明,与其它4种算法相比,在低密度噪声时,该算法的峰值信噪比(Peak Signal-to-Noise Ratio, PSNR)和结构相似性指数(Structural Similarity Index, SSIM)优势明显,在80%高密度噪声时仍然保持突出效果,PSNR值至少提高4.26 dB,验证了该算法对高密度椒盐噪声具有更好的滤波性能,很好地保持图像的纹理边缘和细节。Aiming at the shortcomings of traditional mean filtering and multiple median filtering that cannot effectively remove high-density salt and pepper noise,an iterative adaptive weighted mean filtering algorithm is proposed in this paper.In order to filter the peripheral pixels of the image,the image is first expanded along the surrounding areas.According to the extreme characteristics of salt and pepper noise,pixels are divided into noise points and signal points.For noisy pixels,multiple iterations are used to fully utilize the results of the previous filtering.The weighted average of the grayscale values of non-noisy points in the filtering window is calculated,and the weighting coefficient is calculated based on the spatial distance from the center pixel of the window;the original values of non-noise signal points is kept.The filtering window size adaptively increases from small to large based on whether there are non-noise pixels in the window.In order to verify the effectiveness of the algorithm,five different algorithms are selected for simulation and comparison within the noise density range of 20%to 80%,and evaluation and analysis are conducted from both subjective and objective perspectives.The simulation results show that compared with the other four algorithms,this algorithm has significant advantages in peak signal-to-noise ratio(PSNR)and structural similarity index(SSIM)under low-density noise,and still maintains prominent effects at 80%high-density noise.The PSNR value is at least 4.26 dB higher than other methods.Visual observation also verifies that this algorithm has better filtering performance against high-density salt and pepper noise,and well preserves the texture edges and details of the image.

关 键 词：椒盐噪声 图像扩展 均值滤波 中值滤波 迭代自适应加权均值滤波 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]